Open AccessA note on the modeling of the effects of experimental time in psycholinguistic experiments
Author(s) -
R. Harald Baayen,
Matteo Fasiolo,
Simon N. Wood,
YuYing Chuang
Publication year - 2022
Publication title -
the mental lexicon
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.457
H-Index - 25
eISSN - 1871-1375
pISSN - 1871-1340
DOI - 10.1075/ml.21012.baa
Subject(s) - factorial , computer science , factorial experiment , mixed model , point (geometry) , subject (documents) , series (stratigraphy) , r package , statistics , mathematics , machine learning , mathematical analysis , paleontology , geometry , library science , biology , computational science
Thul et al. (2020) called attention to problems that arise whenchronometric experiments implementing specific factorial designs are analysed with the generalized additive mixed model (GAMM),using factor smooths to capture trial-to-trial dependencies. From a series of simulations incorporating such dependencies, theyconclude that GAMMs are inappropriate for between-subject designs. They argue that in addition GAMMs come with too many modelingpossibilities, and advise using the linear mixed model (LMM) instead. Asclarified by the title of their paper, their conclusion is: “Using GAMMs to model trial-by-trial fluctuations in experimentaldata: More risks but hardly any benefit”. We address the questions raised by Thul et al. (2020) , who clearlydemonstrated that problems can indeed arise when using factor smooths in combination with factorial designs. We show that theproblem does not arise when using by-smooths. Furthermore, we have traced a bug in the implementation of factor smooths in the mgcv package, which will have been removed from version 1.8–36 onwards. To illustrate that GAMMs now produce correct estimates, we report simulation studies implementing differentby-subject longitudinal effects. The maximal LMM emerges as slightly conservative compared to GAMMs, and GAMMs provide estimatedcoefficients that can be less variable across simulation runs. We also discuss two datasets where time-varying effects interactwith numerical predictors in a theoretically informative way.